Motion trajectory recognition using local temporal self-similarities

Motion trajectories provide a meaningful clue in motion characterization of humans, robots, and moving objects. This paper addresses motion trajectory recognition by exploring local self-similarities of motion trajectories over time. Such temporal self-similarities within a motion trajectory are observed by building a Self-Similarity Matrix (SSM) based on the sigmoid distances between all pairs of points along the motion trajectory. On analysis of SSMs, we develop a self-similarity descriptor that captures the layout of local temporal similarities within a motion trajectory. Such descriptors exhibit a noise stability and invariance to group transformations. Temporal pyramid ordering is used in the BoF approach to quantize a set of self-similarity descriptors as a histogram of visual words, forming a temporal pyramid representation accordingly as input data used for recognition. Our method for recognizing motion trajectories is validated on a sign language dataset. It shows similar or superior performance in comparison with other methods. In particular, a significant improvement in recognition efficiency and robustness to noise are achieved using our method.